Action fails after "React API Cutover": empty job ID → polls /api/v1/jobs/ → "Response is not valid JSON" until timeout

[pierzchala-m]

Summary

After the recent commit 0d8f90d ("React API Cutover ahead of Opensource PR Merge"), mieweb/launchpad@main fails for a previously-working pull_request preview deploy. The container is reported "created" but with an empty job ID, after which Phase 2 polls GET /api/v1/jobs/ (no id), receives a non-JSON body (HTTP 200), logs Response is not valid JSON — retrying 30 times, and then fails with Timed out waiting for job to complete after 300s.

Consuming workflow: mieweb/FHIRTogether PR #51 preview deploy.

Observed logs (from the action, secrets redacted)

Querying API: ***/api/v1/sites/1/containers?hostname=mieweb-fhirtogether-feat-systemd-deployment
--- RAW API RESPONSE ---
{"data":[]}
------------------------
Creating container...
--- Create Payload ---
{ "hostname": "mieweb-fhirtogether-feat-systemd-deployment", "template": "ghcr.io/mieweb/fhirtogether:pr-51", "services": [ ... ] }
----------------------
Container created successfully (ID: , Job ID: ).      <-- both empty
...
Attempt 1/30 — HTTP 200
    Response is not valid JSON — retrying.
...
Attempt 30/30 — HTTP 200
    Response is not valid JSON — retrying.
::error::Timed out waiting for job to complete after 300s.

Root cause (response-schema drift)

The backend's JSON schema changed in the React cutover, but action.yml's parsing wasn't fully updated to match:

1. List endpoint now returns {"data":[ ... ]}, but the action reads .containers[0].id:

   GET /api/v1/sites/$SITE_ID/containers?hostname=...   ->  {"data":[]}
   CONTAINER_ID=$(echo "$EXISTING_CONTAINER" | jq -r '.containers[0].id // empty')
   

   .containers no longer exists, so this is always empty.

2. Create response no longer exposes .container.id / .id / .jobId where the action looks:

   NEW_ID=$(echo "$CREATE_BODY" | jq -r '.container.id // .id // empty')
   JOB_ID=$(echo "$CREATE_BODY" | jq -r '.jobId // empty')
   

   Result: Container created successfully (ID: , Job ID: ). — both empty.

3. Empty JOB_ID then makes Phase 2 poll GET /api/v1/jobs/ (trailing-empty id). With no job id, the API returns the SPA HTML page (HTTP 200), so jq empty rejects it as non-JSON on every one of the 30 attempts, then the step times out.

So the failure isn't a transient outage — it's a consistent schema mismatch introduced by the API cutover while @main is consumed mid-migration.

Suggested fixes

• Update action.yml parsing to the new schema (e.g. read the list from .data[...] and the create response/job id from their new locations), or
• Have the API keep returning the legacy keys (containers, container.id, jobId) until the action is updated, and
• Defensive: if JOB_ID is empty after create, fail fast with a clear error instead of polling /api/v1/jobs/ (which returns HTML and wastes 5 minutes). Optionally validate that the jobs endpoint returns application/json before parsing.

Environment

• Action ref: mieweb/launchpad@main (failing commit ≈ 0d8f90d)
• Trigger: pull_request (opened/synchronize)
• Endpoints in use: /api/v1/sites/1/containers, /api/v1/jobs/<id>

[pierzchala-m]

Failed actions:
https://github.com/mieweb/FHIRTogether/actions/runs/27348366384/job/80803597853

https://github.com/mieweb/FHIRTogether/actions/runs/27349339410/job/80806539592